Zero to launch: AI-powered campaign creation without the creative logjam

At the November MarTech Conference, Molly St. Louis, co-founder of Mosaic Group Media, led a session on building and scaling AI-assisted content operations — without losing control of brand, process, or risk. The panelists for the discussion were A. Lee Judge, co-founder and CMO, Content Monsta; Angela Vega, director of capabilities and operations, Expedia Group; and Eric Mayhew, co-founder, president and CPO, Fluency.

The quick hits: Tools that actually earn their keep

• Napkin (Mayhew): For sketching ideas visually when you’re “not artistic.”
• OpenAI platform (Judge): As the backbone for research, coding and automation — but watch model freshness.
• Replit (Vega): To turn product requirements into demo-ready experiences, and Abacus AI to compare outputs from Gemini, Claude and OpenAI.

Tip: Panelists warned many “shiny” apps are just UIs on older models. If results sound dated, check which LLM version it’s running.

What’s really blocking adoption

A live poll showed familiar barriers: tool sprawl (30%), legal fears, internal red tape, and team resistance. Panelists tied these to risk tolerance and ownership confusion—“who writes the check when a machine gets it wrong?” The fix: define governance and spell out where AI is and is not trusted.

Strategy before software

1. Map risk tolerance.
Mayhew urged teams to locate low-risk zones—ideation, variant generation—and pilot there first.

2. Redesign around jobs-to-be-done.
Vega advised deconstructing workflows. “Our old processes were built for yesterday’s tech,” she said. Focus on outcomes—clarity, compliance, channel fit—then re-compose steps with AI plus humans.

3. Start with human input.
Judge: “AI scales what people already said. If you skip the human step, you’ll just repeat what’s out there.”

Brand consistency at scale

Vega described an enterprise playbook:

• Centralize context. Use a Model Context Protocol (MCP) or RAG layer so all agents pull the same brand standards and voice guidelines once, reducing token cost and drift.
• Layer channel rules. One brand core, separate “skills” for email, social or PR.
• Protect data. As AI moves into production, privacy and no-data-egress setups (e.g., Bedrock-style deployments) become must-haves.

Judge applies the same logic to multi-brand work: build discrete voice portfolios for each client or product line so every output “sounds right.” The same library powers video scripts, converting white papers into outlines that match tone and terminology.

From pilot to scale

Early stage:

• Pick one workflow (e.g., ad-copy variants).
• Measure time saved and approval rates.
• Add review gates before public release.

Advanced stage:

• Automate next-best-action hand-offs between nurture and sales.
• Use an orchestration layer to route tasks across models and maintain version history.
• Log prompts, reviewers, and outcomes for auditability.

The human/AI contract

• Treat AI like a coworker, not a calculator.
  Mayhew reframed his mindset: “You’re having a conversation, not issuing code.” Verify facts afterward, but engage it as a creative partner.
• Right-sized context matters.
  Vega: “Too little and it’s lost; too much and it’s noise. Give it the same context you’d give a colleague.”
• Lead with originality.
  Judge: “Pure AI copy lacks novelty. Start from human ideas if you want to rise above the noise.”

Governance, cost and when to pay

Move from free to paid once features or privacy safeguards deliver clear ROI. Evaluate:

1. Interoperability (APIs, webhooks).
2. Community and support.
3. Transparency on model versions and data use.

Centralizing brand context also trims token costs by preventing redundant prompts.

When the issue isn’t AI — it’s process

An audience member asked how to track work across fast campaigns. The answer: combine AI orchestration with disciplined foldering and PM systems. Judge keeps each client’s automations, files, and project boards mirrored across tools — an old-school process that AI can then amplify.

Key takeaways

1. Design around outcomes, not old steps.
2. Centralize brand data and context.
3. Start human, scale with AI.
4. Pilot in low-risk zones, then expand.
5. Measure, version, and govern like it’s production.